How I Kept AI Image Generation Costs Near Zero On A 2 GB Single-Core Server

Описание и применение

Пайплайн генерации изображений на VPS с 2 ГБ ОЗУ и одним ядром, использующий последовательную обработку, сборку метаданных, контроль пиковой памяти и упаковку конечных продуктов, позволяет держать расходы на API почти нулевыми.

Задача

Running an AI image generation pipeline on a 2 GB single‑core VPS caused frequent out‑of‑memory crashes and wasted API credits due to repeated regenerations and metadata rejections.

Результат

The pipeline processed 39 finished products, producing ~856 MB of output from ~19,000 lines of code, while keeping API costs near zero by eliminating duplicate uploads and metadata rejections.

Рабочий процесс

• 1. Maintain an upload ledger (upload_tracker.py) to skip already submitted assets.
• 2. Generate platform‑specific CSV headers with make_metadata.py, optionally using titles.json.
• 3. Process images sequentially on the single core, calling gc.collect() after each asset.
• 4. Measure peak RSS with resource.getrusage(RUSAGE_SELF).ru_maxrss after each run.
• 5. Render heavy steps (e.g., image resizing) in a child subprocess to reclaim memory.
• 6. Render at the required resolution (e.g., 60 dpi for previews, 300 dpi for final).
• 7. Package finished products with pack_product.py, creating a cover (1280×720) and thumbnail (600×600) using Pillow.
• 8. Convert Markdown to styled PDF with md2pdf.py using reportlab.
• 9. Store all research and logs in a permanent research/ folder to avoid re‑doing work.

Доказательства

I killed a batch job at 2 AM because the server had 40 MB of RAM left and I could hear the fan spinning like it was trying to escape.

Источник

When people talk about AI image costs they mean API credits. My costs stayed near zero for a different reason: I stopped throwing away work.

Источник

All three big channels reject a wrong CSV header instantly.

Источник

Four habits, all of them learned by hitting a wall first.

Источник